multi-model-councillisted

# multi-model-council **The problem this solves:** a single model gives you one answer, states it with total confidence, and you're the only thing standing between a confident mistake and your client. Most of the time it's right. But on the work that's *expensive* to get wrong — the memo that reprices a retainer, the positioning call, the audit a client will act on — "most of the time" isn't good enough, and when you're busy, nothing catches the miss. You become the sole reviewer exactly when you have the least attention to spare. `multi-model-council` fixes that by refusing to trust a single answer on work that matters. You send the same task to **two or three different models**, each judges it independently **against a rubric you wrote**, the council **surfaces where they disagree**, and you take the version they converge on — majority rules, or you break the tie with the disagreement laid out in front of you. It's the cheapest insurance you can buy on high-stakes work: a second and third opinion that actively pick each other apart catch the fluent, confident mistake before it ever leaves your desk. And once it's working for you, it's the moment the *team* goes agentic — the quality bar stops depending on you being the one who looks. > The plumbing that reaches several models lives in `cli-llm-routing` — the skill that routes a task to different model CLIs for a second opinion or a skeptical review. `multi-model-council` is the **pattern** built on top of it: not just "a